TL;DR
MiniMax has launched M3.1 Flash, a model the company says is designed for coding, agentic workflows and computer-use tasks. MiniMax lists API prices of $0.15 per million input tokens and $0.60 per million output tokens, and claims output speeds can exceed 200 tokens per second; its benchmark results and performance claims have not been independently verified in the supplied material.
MiniMax has launched M3.1 Flash, an AI model aimed at coding, agentic workflows and computer use, and says it is available through the MiniMax API, M Plan and MiniMax Code. The company is promoting lower prices and faster inference, but the performance figures and comparisons in its launch report are MiniMax’s own claims; the supplied material does not include independent evaluations.
Get business pricing on tech for your team
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
MiniMax describes M3.1 Flash as a next-generation Flash model built for tasks in which AI systems use tools or carry out multi-step work. The company says it improves on its predecessor across automation, software development, computer operation and cybersecurity. It also says an earlier version operated anonymously as “Space Bunny” and processed more than 200 trillion tokens. That figure and the identity of the earlier model are reported by MiniMax, with no independent corroboration supplied.
The launch report lists several benchmark results: 76.2 on BankerToolBench, 73.7% on DeepSWE v1.1, 61.7 on OSWorld V2, and 29.6 on Agents’ Last Exam, comparisons also discussed in industry coverage of DeepSeek Flash. MiniMax says the OSWorld V2 and ALE results rank first among Chinese models, and reports an 82% CyberGym score. The company presents the results as evidence of performance in enterprise workflows, coding, computer use and security testing; the report does not provide outside verification or enough testing detail to assess those comparisons independently.
MiniMax lists prices of $0.15 per million input tokens, $0.60 per million output tokens, and $0.015 per million cached input tokens. It says agentic-task usage costs are 70% below those of its predecessor. On speed, the company reports median output above 170 tokens per second and says everyday output can reach more than 200 tokens per second. These are vendor-reported figures, and the announcement does not specify a testing setup, hardware conditions or workload for the speed claims.
Lower Prices for Tool-Using AI
The release targets a practical barrier to using AI in production: the cost and latency of models asked to perform long, tool-heavy tasks. If MiniMax’s figures hold in customers’ workloads, lower token prices and faster output could make it less expensive to run coding assistants, automate routine business processes or connect models to computer interfaces. Those applications can consume many model calls as they plan, act and check results, so per-token pricing alone may not capture total operating cost.
The announcement is also part of competition among model providers to offer capable systems at lower prices. MiniMax frames M3.1 Flash as a balance between capability and efficiency, while saying a separate Pro model is planned. For buyers, the central question is not only whether the model scores well on selected tests, but whether it reliably handles their own tasks, integrates with their tools, and delivers savings after retries, supervision and infrastructure costs are included.
Products Worth Considering
DJI Osmo Pocket 3 Vlogging Camera with 1-Inch CMOS Sensor and 4K/120fps Video
Samsung 27-inch Odyssey OLED G6 QHD Gaming Monitor
Logitech MX Master 4 Ergonomic Wireless Mouse with Haptic Feedback, Graphite
As an affiliate, we earn on qualifying purchases.
Availability, Pricing and Planned Weights
MiniMax says customers can access M3.1 Flash through three routes: the MiniMax API, its M Plan subscription, and MiniMax Code. The launch report links to the company’s API platform, subscription page and desktop code product. It does not give a calendar date in the provided text, so the announcement’s “today” cannot be tied to a specific date here.
The model’s weights are not yet available, according to MiniMax, which says they will be released later. That distinction matters for developers choosing between hosted access and running or adapting a model themselves. MiniMax also says its Flash line explores efficiency and capability, and that a Pro model is expected later, without giving a launch date or further specifications.
““Today, we officially roll out MiniMax M3.1 Flash.””
— MiniMax
Independent Tests Still Needed
The supplied material is a company launch report, not an independent evaluation. It does not identify benchmark versions and settings in enough detail to reproduce every result, name comparison systems, or provide third-party testing. MiniMax’s claims that the model performs at the level of leading models and ranks first among Chinese models should therefore be read as vendor-reported comparisons, not independently established rankings.
It is also unclear how the quoted speed and cost reductions translate to particular customer workloads. The announcement does not specify hardware, prompt lengths, concurrency, output lengths or the baseline used for the 70% cost comparison. It gives no schedule for the model-weight release or the planned Pro model. The claims about the earlier “Space Bunny” version processing more than 200 trillion tokens are likewise not supported with usage records or further detail in the supplied source.
Weights and Pro Model Await Dates
Developers can try the hosted model through the API, M Plan or MiniMax Code, according to the announcement. Those routes provide an immediate way to test its prices and performance against specific tasks, though results will depend on each workload and implementation. MiniMax says it plans to release the model weights later, which could broaden access beyond the current products, but it has not announced a date.
The company has also said a Pro model is coming, without providing a timetable or technical details. Until more information is released, the main near-term questions are whether independent testers reproduce the benchmark and speed claims, how well the system performs in real deployments, and when weights and the Pro model will become available.
Key Questions
What is MiniMax M3.1 Flash?
It is a MiniMax AI model promoted for agentic tasks, coding and computer use. The company also reports cybersecurity benchmark results and describes the model as an efficiency-focused release.
Where can people access M3.1 Flash?
MiniMax says it is available through the MiniMax API, M Plan and MiniMax Code. The launch report says model weights will come later.
How much does the model cost?
MiniMax lists prices of $0.15 per million input tokens, $0.60 per million output tokens and $0.015 per million cached input tokens. These are the company’s published rates in the supplied announcement.
Are the benchmark and speed claims independently verified?
Not in the supplied material. The reported scores, rankings and speeds come from MiniMax’s launch report; no independent evaluation or testing conditions are included there.
When will the model weights and Pro model arrive?
MiniMax says the weights will be released later and that a Pro model will also arrive later, but it gives no dates or additional specifications in the announcement.
Source: hn
